diff --git a/apps/site/docs/en/API.mdx b/apps/site/docs/en/API.mdx
index 82f2d99d2..d54cae6ba 100644
--- a/apps/site/docs/en/API.mdx
+++ b/apps/site/docs/en/API.mdx
@@ -394,7 +394,6 @@ function aiAssert(assertion: string, errorMsg?: string, options?: {
   * `errorMsg?: string` - An optional error message to append if the assertion fails.
   * `options?: Object` - Optional, a configuration object containing:
     * `deepThink?: boolean` - If true, Midscene will enable AI model's [COT](https://www.wikiwand.com/en/articles/Chain_of_thought) mode, which makes AI model to think more deeply when asserting. Default: False.
-    * `keepRawResponse?: boolean` - If true, Midscene will keep the raw response from the AI model. Default: False.
 
 * Return Value:
   * Returns a Promise that resolves to void if the assertion passes; if it fails, an error is thrown with `errorMsg` and additional AI-provided information.
diff --git a/apps/site/docs/zh/API.mdx b/apps/site/docs/zh/API.mdx
index 5d8ce8c96..968c5db46 100644
--- a/apps/site/docs/zh/API.mdx
+++ b/apps/site/docs/zh/API.mdx
@@ -397,7 +397,6 @@ function aiAssert(assertion: string, errorMsg?: string, options?: {
   * errorMsg?: string - 当断言失败时附加的可选错误提示信息。
   * options?: Object - 可选的配置对象，包含：
     * `deepThink?: boolean` - 是否开启深度思考。如果为 true，Midscene 启用 AI 模型的 [COT](https://www.wikiwand.com/en/articles/Chain_of_thought) 模式，使 AI 在断言时进行更深入的思考。默认值为 False。
-    * `keepRawResponse?: boolean` - 是否保留 AI 的原始响应。如果为 true，Midscene 会保留 AI 的原始响应而不格式化。默认值为 False。
 
 * 返回值：
   * 返回一个 Promise。当断言成功时解析为 void；若断言失败，则抛出一个错误，错误信息包含 `errorMsg` 以及 AI 生成的原因。
diff --git a/packages/core/src/ai-model/prompt/assertion.ts b/packages/core/src/ai-model/prompt/assertion.ts
index 9eea1d3de..71f600f29 100644
--- a/packages/core/src/ai-model/prompt/assertion.ts
+++ b/packages/core/src/ai-model/prompt/assertion.ts
@@ -7,17 +7,35 @@ const defaultAssertionPrompt =
 const getDefaultAssertionResponseJsonFormat = (
   deepThink: boolean,
 ) => `Return in the following JSON format:
-{
-  thought: string, // string, ${deepThink ? 'ALWAYS provide the reasoning process that led to the pass/fail conclusion. This should detail the step-by-step thinking.' : 'if the result is falsy, give the reason why it is falsy. Otherwise, put null.'}
+${
+  deepThink
+    ? `{
+  thought: string, // string, ALWAYS provide the reasoning process that led to the pass/fail conclusion. This should detail the step-by-step thinking.
   pass: boolean, // whether the assertion is truthy
-}`;
+}`
+    : `{
+  pass: boolean, // whether the assertion is truthy
+  thought: string, // string, if the result is falsy, give the reason why it is falsy. Otherwise, put null.
+  }`
+}
+`;
 
-const getUiTarsAssertionResponseJsonFormat = () => `## Output Json String Format
+const getUiTarsAssertionResponseJsonFormat = (
+  deepThink: boolean,
+) => `## Output Json String Format
 \`\`\`
-"{
-  "thought": "<<is a string, give the reason why the assertion is falsy or truthy. Otherwise.>>"
+${
+  deepThink
+    ? `{
+  "thought": "<<is a string, ALWAYS provide the reasoning process that led to the pass/fail conclusion. This should detail the step-by-step thinking.>>", 
   "pass": <<is a boolean value from the enum [true, false], true means the assertion is truthy>>, 
-}"
+}`
+    : `{
+  "pass": <<is a boolean value from the enum [true, false], true means the assertion is truthy>>, 
+  "thought": "<<is a string, give the reason why the assertion is falsy or truthy. Otherwise.>>"
+
+}`
+}
 \`\`\`
 
 ## Rules **MUST** follow
@@ -31,7 +49,7 @@ export function systemPromptToAssert(model: {
 }) {
   return `${defaultAssertionPrompt}
 
-${model.isUITars ? getUiTarsAssertionResponseJsonFormat() : getDefaultAssertionResponseJsonFormat(model.deepThink)}`;
+${model.isUITars ? getUiTarsAssertionResponseJsonFormat(model.deepThink) : getDefaultAssertionResponseJsonFormat(model.deepThink)}`;
 }
 
 export const assertSchema: ResponseFormatJSONSchema = {
diff --git a/packages/core/tests/unit-test/prompt/__snapshots__/assertion.test.ts.snap b/packages/core/tests/unit-test/prompt/__snapshots__/assertion.test.ts.snap
index 4dc90b1ef..a3b62d2b5 100644
--- a/packages/core/tests/unit-test/prompt/__snapshots__/assertion.test.ts.snap
+++ b/packages/core/tests/unit-test/prompt/__snapshots__/assertion.test.ts.snap
@@ -5,10 +5,11 @@ exports[`Assertion prompt > return UI-Tars specific when it is UI-Tars 1`] = `
 
 ## Output Json String Format
 \`\`\`
-"{
-  "thought": "<<is a string, give the reason why the assertion is falsy or truthy. Otherwise.>>"
+{
   "pass": <<is a boolean value from the enum [true, false], true means the assertion is truthy>>, 
-}"
+  "thought": "<<is a string, give the reason why the assertion is falsy or truthy. Otherwise.>>"
+
+}
 \`\`\`
 
 ## Rules **MUST** follow
@@ -22,10 +23,10 @@ exports[`Assertion prompt > return UI-Tars specific when it is UI-Tars and deepT
 
 ## Output Json String Format
 \`\`\`
-"{
-  "thought": "<<is a string, give the reason why the assertion is falsy or truthy. Otherwise.>>"
+{
+  "thought": "<<is a string, ALWAYS provide the reasoning process that led to the pass/fail conclusion. This should detail the step-by-step thinking.>>", 
   "pass": <<is a boolean value from the enum [true, false], true means the assertion is truthy>>, 
-}"
+}
 \`\`\`
 
 ## Rules **MUST** follow
@@ -39,7 +40,8 @@ exports[`Assertion prompt > return default when it is not UI-Tars 1`] = `
 
 Return in the following JSON format:
 {
-  thought: string, // string, if the result is falsy, give the reason why it is falsy. Otherwise, put null.
   pass: boolean, // whether the assertion is truthy
-}"
+  thought: string, // string, if the result is falsy, give the reason why it is falsy. Otherwise, put null.
+  }
+"
 `;